Techniques for parsing code blocks without curly braces

Question

I'm writing a simple parser/interpreter in C# from scratch (no third-party libraries). It compiles to bytecode and then I have class that runs the bytecode. I'm getting close to wrapping it up. I've just implemented while and for loops and am working on if|else if|else blocks.

As it stands, my parser requires all of these structures to use curly braces. I'd like to make it more C-like and have the curly braces be optional when the block contains just a single statement. This is giving me trouble.

if (condition)
{
    // Make curly braces optional when there is just one statement here
}

The problem is tracking state. How does the parser know when a block without curly braces has ended. One approach would be to check if there is a block without braces in effect after each and every statement. However, there are a lot of different scenarios that would constitute a statement and so those checks would need to be in a number of places. That feels a little brittle to me.

I'm just wondering if anyone has done this and knows of any slick tricks for tracking when a code block ends when there are no curly braces.

Answer 1

You need to look into recursive descent parser. It makes creating parsers a lot easier. Lets assume you have grammar looking like this:

statement
   : 'if' paren_expr ['{'] statement ['}'] 

paren_expr
   : '(' expr ')'

then using recursive descent you can do something like:

public void Statement()
{
    if(curToken == Token.If)
    {
       Eat(Token.If); // Eat is convenience method that moves token pointer on
       if(curToken == Token.LParen)
       {
          Eat(Token.LParen)
          ParenExpr();
          Eat(Token.RParen);
       }
       if(curToken == Token.LBrace) // this will signify a block of statements
       {   
          Eat(Token.LBrace);
          while(curToken != Token.RBrace)
             Statement();
          Eat(Token.RBrace);
       }
       else
          Statement();              
    }
}

public void ParenExpr()
{
   // do other token checks
}

doing this for all of your non terminals, you can easily build up an AST and from that, you can generate your bytecode.